Two of the most common ways to hear hockey observers describe a goalie’s performance in a single game are “giving the team a chance to win” or “stealing the game.” The former refers to playing well enough for the team to stay in the game and have a chance to get a win. The latter refers to playing so well that the team is able to win a game in which its skaters were outplayed by the opposition.
These descriptions are typically used based on the flow of the game or on the number of goals allowed. They are subjective and can hold different meanings in different contexts. Sometimes giving the team a chance to win means holding down the fort after allowing two early goals to allow for a comeback. Sometimes it means giving up five goals in a 6-5 win but never allowing the other team to take the lead. Steals are a little more specific and usually refer to a game where the goalie played exceptionally well under heavy pressure and made a low scoring effort from their own team good enough to win.
But what if instead of leaving these concepts to the realm of color commentators making quick assessments of goalies on the fly, we could put some parameters around the definitions of these terms. The drawback of that no matter what definition we use, it will be wrong to a certain extent and never capture the exact essence of the nebulous concept it attempts to define. That’s inevitable when trying to turn a feeling into a number. But the advantage is that if we create a numeric version of that feeling, we can count it. And if we can count it, we can see who gives their team a chance to win most often and who steals games most often. That’s fun!
So, while acknowledging that our definition will never be perfect, let’s try to put some walls around this free flowing language. All the data used to calculate the numbers in this article is via Evolving Hockey. Specifically, we’re using their play by play query tool to gather the information we need. To get access to that, subscribe to their Patreon.
Let’s start here because this will be the easier of the two but also the most contentious. Quality Starts in hockey have existed for a long time, at least since 2009 and probably much longer. They’ve existed for pitchers in baseball since 1985 so it wouldn’t be a surprise to find that hockey teams have had an internal notion of this idea going back to the 80s or 90s.
The basic idea of of a Quality Start, in baseball or hockey, is that the (pitcher or) goalie has played well enough to give their team a chance to win. They aren’t necessarily playing excellently. They don’t have to win the game on their own. But they’re playing well enough that if the rest of the players on the ice do their job, the team should win.
The current definition, which is available at Hockey Reference, comes from Rob Vollman who now works for the Los Angeles Kings. It awards a quality start to goalies who post a save percentage at least as high as the league average for that season or if they allow two goals or less and post a save percentage above that of a replacement level goalie.
This was a reasonable definition in 2009 when Vollman created it but we have data available now that wasn’t then. This new data allows for improvements. Nick Mercandante moved in this direction in 2016 when he created Above Average Appearances, which incorporated some information on shot danger by giving goalies credit for games where they posted above average save percentages against low, medium, and high danger shots.
We can take Nick’s approach further but at the same time, simplify it because of the new data available. With consistent access to expected goals, we can define a Quality Start as any game in which the goalie does not allow more goals than would be expected. In other words, any game where a goalie posts a goals saved above expected of at least zero is a Quality Start. Because expected goal models account for shot quality, we don’t have to fuss with separate shot danger levels. This definition is both reasonable and simple, which makes it preferable to the other options.
Of these two concepts, steals is the more fun one. Several people have done previous work in the area of Quality Starts but the only previous work I could find on the idea of stealing games was on Twitter by Cole Anderson. Yet, we hear about it constantly on broadcasts. And rightfully so. Games where the goalie is the reason a team wins are certainly a thing. But what would a definition for this type of game look like?
Again, we can turn to expected goals models to solve this for us. And maybe even more easily than we might expect at first. If a quality start is any time the goalie’s goals saved above expected is at least zero, then I suggest a steal is any time the goalie’s goals saved above expected is greater than the final goal differential in the game. This is a similar definition to the one used by Cole in his work.
For example, let’s say the Lightning beat the Toronto Maple Leafs 3-2. If the Leafs posted an expected goal total of 4.0 in that game, that means Andrei Vasilevskiy saved 2.0 goals above expected. And if he hadn’t done that, the Leafs would have won the game. Had he performed as expected, the Lightning would have lost 4-3. His 2.0 goals saved above expected were greater than the goal differential of 1.0, so he stole the win for his team in that scenario.
That was a lot of words defining things so let’s get to the fun part and look at some charts. The following plot includes the top 45 goalies in the NHL this season in games played. The horizontal axis is Quality Start Percentage and the vertical axis is Steal Percentage. By percentage here, we mean what percentage of their total games played qualify as quality starts or steals. So if a goalie has played 36 games and posted 18 quality starts, that would be a 50% quality start percentage. For context, the dots on the plot are colored by the amount of games played so far this season.
What should be immediately clear from this chart is that these numbers don’t tell us much about who the best goalies are. If we want to know who’s playing the best over a given timeframe, we can look at goals saved above expected. What we’re seeing here is how frequently goalies have a particular type of game that we find interesting.
Petr Mrazek immediately stands out from the pack. He’s both stolen quite a few games this year and posted a high percentage of quality starts. At the other end, Jimmy Howard has rarely performed up to expectations giving up more goals than expected in nearly every start.
If we’re being Lightning specific, Andrei Vasilevskiy is hovering a little bit below average in both measures among this group of goalies. Curtis McElhinney didn’t qualify because he’s only played in 11 games. While he hasn’t stolen any games he has provided six quality starts, which is a higher percentage than Vasilevskiy albeit in a lesser workload both in terms of games played and competition.
While one season is fun to see, we can also pull back further and make the same chart looking at the last five seasons. This version includes the top 45 goalies in games played going back to 2015-2016.
This view aligns a bit more with our understanding of who the best goalies are. John Gibson is near the front of the line in both quality start percentage and steal percentage. Being a great goalie on a terrible team for the past few years has given him lots of opportunities to steal games, which is something that doesn’t happen as frequently for a goalie on a high powered offensive team like the Lightning.
Andrei Vasilevskiy is again well below average among his peer group. And while neither of these measures say anything about overall quality of play, his place on this chart does align with his results in terms of goals saved above expected so far in his career.
We hear talk of goalies giving their team a chance to win or stealing a game all the time. When we have concepts like this that are accepted as part of the discourse around the game, it can be fun to try to define a framework around them so that we can measure them. In this case, the thing we’re counting isn’t important in terms of determining how well a goalie is playing in aggregate.
But these types of stats help describe how a player gets to their results and add some color. As always in hockey, our eyes lie to us all the time. We might have a feeling that a goalie is stealing a lot of games or that they are always playing well enough for the skaters to have a chance to earn a victory. By defining these concepts, we can then check to see if our gut feelings are supported or refuted by a more structured approach.